多标签分类

此示例展示了如何格式化多标签分类问题的目标值。有关多标签分类的详细信息可在此处找到 here。

import numpy as np
from pprint import pprint

import sklearn.datasets
import sklearn.metrics
from sklearn.utils.multiclass import type_of_target

import autosklearn.classification

数据加载

# Using reuters multilabel dataset -- https://www.openml.org/d/40594
X, y = sklearn.datasets.fetch_openml(data_id=40594, return_X_y=True, as_frame=False)

# fetch openml downloads a numpy array with TRUE/FALSE strings. Re-map it to
# integer dtype with ones and zeros
# This is to comply with Scikit-learn requirement:
# "Positive classes are indicated with 1 and negative classes with 0 or -1."
# More information on: https://scikit-learn.cn/stable/modules/multiclass.html
y[y == "TRUE"] = 1
y[y == "FALSE"] = 0
y = y.astype(int)

# Using type of target is a good way to make sure your data
# is properly formatted
print(f"type_of_target={type_of_target(y)}")

X_train, X_test, y_train, y_test = sklearn.model_selection.train_test_split(
    X, y, random_state=1
)

type_of_target=multilabel-indicator

构建分类器

automl = autosklearn.classification.AutoSklearnClassifier(
    time_left_for_this_task=60,
    per_run_time_limit=30,
    # Bellow two flags are provided to speed up calculations
    # Not recommended for a real implementation
    initial_configurations_via_metalearning=0,
    smac_scenario_args={"runcount_limit": 1},
)
automl.fit(X_train, y_train, dataset_name="reuters")

AutoSklearnClassifier(ensemble_class=<class 'autosklearn.ensembles.ensemble_selection.EnsembleSelection'>,
                      initial_configurations_via_metalearning=0,
                      per_run_time_limit=30,
                      smac_scenario_args={'runcount_limit': 1},
                      time_left_for_this_task=60)

查看 auto-sklearn 找到的模型

print(automl.leaderboard())

          rank  ensemble_weight           type      cost  duration
model_id
2            1              1.0  random_forest  0.447294  4.031171

打印 auto-sklearn 构建的最终集成模型

pprint(automl.show_models(), indent=4)

{   2: {   'balancing': Balancing(random_state=1),
           'classifier': <autosklearn.pipeline.components.classification.ClassifierChoice object at 0x7f05d248c2e0>,
           'cost': 0.4472941828699525,
           'data_preprocessor': <autosklearn.pipeline.components.data_preprocessing.DataPreprocessorChoice object at 0x7f05d0f2e520>,
           'ensemble_weight': 1.0,
           'feature_preprocessor': <autosklearn.pipeline.components.feature_preprocessing.FeaturePreprocessorChoice object at 0x7f05d248c8b0>,
           'model_id': 2,
           'rank': 1,
           'sklearn_classifier': RandomForestClassifier(max_features=15, n_estimators=512, n_jobs=1,
                       random_state=1, warm_start=True)}}

打印 auto-sklearn 运行的统计信息

# Print statistics about the auto-sklearn run such as number of
# iterations, number of models failed with a time out.
print(automl.sprint_statistics())

auto-sklearn results:
  Dataset name: reuters
  Metric: f1_macro
  Best validation score: 0.552706
  Number of target algorithm runs: 1
  Number of successful target algorithm runs: 1
  Number of crashed target algorithm runs: 0
  Number of target algorithms that exceeded the time limit: 0
  Number of target algorithms that exceeded the memory limit: 0

获取最终集成模型的得分

predictions = automl.predict(X_test)
print("Accuracy score", sklearn.metrics.accuracy_score(y_test, predictions))

Accuracy score 0.604